Introduction to Bioinformatics

Prep readings:

• Text: Class notes
• Files:
  • featuredet.m
  • findmatches.m
  • markmatches.m
  • ccr5gel.tif
  • target.tif

In this assignment, we wish to:

• learn how to do feature detection on an image by template matching to find a specific target

Consider the noisy images shown above. Your goal is to design a computer program to count the number of blot features in the gel on the left. You begin by extracting a sample blot subimage, such as the one on the right. To find the blots in the original image, you simply match the template image to all (or many) of the possible places it could be located at in the original image. In image processing, this type of algorithm is known by various names, such as template matching or model matching or feature detection. The blot subimage represents the template, model, or feature you are interested in matching or detecting. The matching/detection process can be implemented in several equivalent ways:

• Count the number of agreements between pixels in the image and the template, and then pick the location(s) that have the maximum number of agreements. This is known as a maximum correlation approach.
• Count the number of disagreements between pixels in the image and the template, and then pick the location(s) with the minimum number of disagreements. This is known as a minimum error approach.
• Use a distance function (typically a simple Euclidean distance) applied between the image pixels and the template pixels to measure the similarity of the template and the image at each location. This approach then picks the location(s) with the smallest distance as the location(s) of the template image in the original image.

Getting some practice

• Copy the files and TIF images linked above into your bioimaging folder in your ArgoNet account. (This was the folder you created in Introduction to Imaging in Bioinformatics (Part I).)
• Start MATLAB, and run the template matching program by typing featuredet program in the MATLAB Command Window on the right. The output you see in the command window will indicate the row number of the original image as it is being searched. You will also see a final array indicating the location(s) detected and an associated measure for each location.
• If all works correctly, you should also see a figure open up with the results indicated on the original image. Any results that scored a 100% match are shown in red. Other matches are shown in shades of yellow.
• Study the results. What is the difference between the two result images provided? How many targets were located in each of these images? Which image set has better performance and why? How did this algorithm perform overall on the blot counting task?
• Now, open the findMatches.m file and change the value for the percentMatch variable. (Try both higher and lower values.) Run the program again and observe any results.
• Next, open the featuredet.m file and change the value of the TOTAL_MATCHES variable. (Try both higher and lower values.) Run the program again and observe any results.
• As you complete this lab, think carefully about the following questions:
  • Which of the three methods for matching (described above) is implemented in this code? What are the advantages and disadvantages of this approach?
  • How would the presence of noise impact template matching?
  • How would template matching be impacted by any distortions of the template or targets in the original image, such as translation, rotation, shearing, expansion, contraction or occlusion?
  • What are some ways to improve the speed of template matching?

• If you see the error message:

  ??? LZW-compressed TIFF images are not supported.

  you'll need to turn off compression when saving the images. For example, in Microsoft Photo Editor, you can do this by selecting File-Save As and then choose "More>>" to get to the "Compression" option, which you can turn to "None" in the pull-down menu.
• If you see the error message

  ??? Index exceeds matrix dimensions.

  you'll need to make sure you have saved the images in 24-bit color. In Microsoft Photo Editor, you can do this by selecting File-Save As and then choose "More>>" to get to the "Convert To" option, which you can turn to "True Color (24 bit)" in the pull-down menu.
• If MATLAB is taking too long to process your images, resize them. I recommend starting with original images which are about 50-100 rows. Your target/template image would then be much smaller than this. In Microsoft Photo Editor you can easily resize an image using Image-Resize and providing a percentage. Remember to save your new image. You can also crop images easily with this tool by choosing the Select option (the dashed box in the options at the top of the tool), and then choose Image-Crop-OK, followed by saving.

To get credit for this assignment, you need to complete these additional steps:

1. Find an interesting image that you would like to do some template matching on. Ideally, this might be an image related to your Term Project. Store the new image in TIF format in your bioimaging folder. Next, use one of the Microsoft tools with an image cropping option to extract a feature image. Save this in TIF format as well. (Other image formats such as JPEG that involve compression may not work as well.)
2. Make a copy of the main featuredet.m MATLAB program provided above (also in your bioimaging folder) and name the copy my_featuredet.m
3. In your new program file, change the names of the input images to the images you would like to process. Save this new program, and then type my_featuredet in the MATLAB Command Window to run the code with your new images.
4. Experiment with the code and parameter values until you produce some images which appear interesting or useful.
5. Once you have some interesting images, save them from within the MATLAB program (using the imwrite function). Note that by default, MATLAB will save the images in the directory indicated next to the "Current Directory" label in the main window, so you will need to change this directory to your bioimaging folder before you run the new version of the program.
6. On your course Webpage, under a heading titled, "Lab 7b: Imaging in Bioinformatics", include links to:
   • your new MATLAB files
   • four images: your original image, template image, and two result images with the target locations marked
   • text which describes why you think your results are interesting or useful
7. E-mail me (msutton@uwf.edu) when you have completed the above steps.